Deep learning recognition and control method and device for photovoltaic cell visual sorting

Abstract

The present invention provides a deep learning recognition and control method for photovoltaic cell visual sorting. A defect classification and detection method based on deep learning is integrated with an industrial manufacturing process to have high intelligence. The deep learning replaces a traditional learning method to have a high feature extraction capacity, especially when background interference is serious and defects are difficult to identify by the traditional method. In addition, the deep learning recognition and control device can automatically perform defect feature extraction, the processing speed is improved, and the manual dependence is reduced.

Description

technical field [0001] The present invention relates to the field of solar cell image acquisition, and in particular to a deep learning recognition and control method and device for sorting photovoltaic cells Background technique [0002] In the process of industrial manufacturing, the surface defects of workpieces such as crack defects on the surface of photovoltaic cells; some surface appearance defects of photovoltaic cells such as dirty pieces, broken grids, and scratches; surface defects of strip steel have a very important impact on industrial production efficiency . The classification and detection of surface defects play an important role in improving the process, improving the quality of components, improving efficiency and stabilizing production. [0003] Existing sorting devices for photovoltaic cells, such as device CN 108010864 A (a sorting device for defects in photovoltaic cells and its sorting method), have the following problems: (1) It is impossible to ana...

AI Technical Summary

Problems solved by technology

[0003] Existing sorting devices for photovoltaic cells, such as device CN 108010864 A (a sorting device for defects in photovoltaic cells and its sorting method), have the following problems: (1) It is impossible to analyze small-scale defects: Under the interference of grid lines and lattices in the image, small defect images cannot be identified

(2) There are strict requirements on image quality for the defect shape: if the defect part overlaps with the background, or the image quality is low, traditional vision cannot detect such a defect

(4) The processing program is complex and the processing speed is slow: traditional visual inspection methods rely on manual feature extraction, and high accuracy depends on high-complexity programs, resulting in reduced processing speed

Images